D ANIEL L EIVANT
Failure of completeness properties of intuitionistic predicate logic for constructive models
Annales scientifiques de l’Université de Clermont-Ferrand 2, tome 60, série Mathéma- tiques, n
o13 (1976), p. 93-107
<http://www.numdam.org/item?id=ASCFM_1976__60_13_93_0>
© Université de Clermont-Ferrand 2, 1976, tous droits réservés.
L’accès aux archives de la revue « Annales scientifiques de l’Université de Clermont- Ferrand 2 » implique l’accord avec les conditions générales d’utilisation (http://www.
numdam.org/conditions). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
FAILURE OF COMPLETENESS PROPERTIES OF INTUITIONISTIC PREDICATE LOGIC
FOR CONSTRUCTIVE MODELS Daniel LEIVANT
University of Amsterdam, AMSTERDAM, Netherlands
ABSTRACT
We consider a
principle
ofconstructivity
RED which states that every decidablepredicate
over the natural numbers is
(weakly) recursively
enumerable(r.e.).
RED iseasily
seen to bederived from Church’s thesis
CT 0 (every
construction isgiven by
a recursivefunction»).
Results :
(1)
REDimplies
that thespecies
of valid first orderpredicate
schemata is not r.e., andhence - that intuitionistic first order
predicate logic .t 1 is incomplete.
(2)
We construct aspecific
schema of£1
which is valid if RED , butunprovable (3)
The two results above hold even whenvalidity
isgeneralized
tovalidity
withKreisel-Troelstra
[ 70 ]’s
choice sequences asparameters.
(4)
The method is used also to construct a schemaof 9, 1 , unprovable
in L1, but
ofwhose all metasubstitutions
with L 0
number theoreticpredicates
areprovable
in1 p
Heyting’s
arithmetic G. This is asimple
bound onpossible improvements
of themaximality
result of Leivant[ 75 ] .
1. INTRODUCTION
For a
general
introduction to thesubject
matter of this note the reader is referred to Troelstra[76 ] .
’
The main two results of this note are the
following. Firstly,
we show(along
the line ofKreisel
[70] ] p. 133)
that the schema RED mentioned above in the abstractimplies
that thespecies
ofintuitionistically
valid formulae ofpredicate logic
is not r.e.. This we doby exhibiting
a sequence( F )
offormulae,
thevalidity
ofFm~n meaning intuitively :
othe r.e.
species Wm and Wn
are notseparable by
anypair
of decidablespecies». By
REDdecidable
species
are(weakly)
r.e., so we may say insteadWm
andWn
arerecursively inseparable».
Theproof
is then concludedby showing (intuitionistically)
that thespecies
((m,n) ~ I Wm and Wn
arerecursively inseparable)
is not r.e..Secondly,
wepick
up m,n for whichFm n
is notvalid,
but s.t. the recursiveinseparability
ofWm
andWn
isprovable
in intuitionistic arithmetic. This lastproof
may bereshaped
toapply
to anygiven 10
model ofarithmetic ;
we conclude that any1
10
-metasubstitution ofF m n
isprovable
inarithmetic,
thusproving
result(4)
of theI ’
abstract.
1.1. NOTATIONAL CONVENTIONS.
We use the
logical
constants&,
A , -~ ,rr , rq
and .1(for absurdity ; negation
-,is then
definable).
The lettersP,Q,T,Z,S...
are used forpredicate parameters (in
thelanguage
of first-order
predicate logic 1),
while bold face T,Z,S... are used for certain number theoreticpredicates explicitely
defined below. ’For recursion theoretic notions we use the notations of Kleene
[ 52 ]
and[69 ]
: Tn for thecomputation predicate
for n-aryfunctions,
U for theresult-extracting
function,~ e~
for the e’thpartial
recursivefunction, "
forpartial equality,
etc.We assume
given
an encodement of sequencesi.e. -
(n~,
...,nk) 2013~ (
no, ...,nk) together
with alength-function
lth and inverseprojection
functions(n) Iwhich satisfy
We also fix some
G6del-coding
forsyntactic objects,
and write ’ a ’ for the code of a .For formal r.e. theories
ct
we writeProv (p,r)
for a canonicalproof predicate
forCc ,
and
Pr (r)
for 7 p Prov(p,r).
1.2. -1.4. FORMAL SETTING.
1.2. LANGUAGE.
°
Our metamathematical
setting
is the intuitionistictheory
ofspecies, ~ 2.
Aninspection
on the
proofs
below shows thatactually only
a smallfragment
isused,
but since it is of little value for our purpose to delimit thisfragment precisely
we do not bother to do so.Prawitz
( [ 65 ]
p.72)
has shown thatHeyting’s Arithmetic Ci-
isinterpretable
however,
to avoid notationalcomplications
we shall use lower casea,b,c
etc. for numericvariables,
and lower case x,y,xi etc. for unrestricted first order variables. A list zo, ..., zi, ... of unrestrictedvariables will be reserved for a
specific
use, and we shall assume thatthey
do not occur informulae otherwise. We also assume that G as well
as L2
containsymbols
anddefining equations
for allprimitive
recursive functions.The
system
westudy
is first order intuitionisticpredicate logic .9
l’ withoutequality
and without function
symbols.
We shall find it convenient to refer also to extended to thelanguage
with all constants of G( = , numerals
andprim.
rec.functions).
We write1 G
forthe
resulting system.
1.4. THE VALIDITY PREDICATE.
Tarski’s definition of a
II 1
1validity predicate
Val for £1(analogously
1 to the classicalcase)
is standard(cf.
e.g. Tarski[36 J ).
The centralproperty
of Val is of coursefor each schema G where P is the list of all
predicate parameters occurring
inG, D 1
is a(unary) predicate
variable andGD
comes from Gby restricting
allquantifiers
to D.To avoid confusion one should note that the
predicate
Val does not express anyanalysis
of the notion of constructive truth(contrary
to truth definitions like Beth’s andKripke’s
semantics and the variousrealizability predicates).
Here the constructive content of! 1
showsonly through
the constructivemeaning given (in S- 2)
to thelogical
constantsoccurring
in Val. Our discussion is therefore validregardless
of anyspecific analysis
ofconstructivity.
1.4, THE SCHEMA RED.
We define the schema RED, for orecursive
enumerability
of decidablepredicates»
to beIt is
easily
seen that RED is derivable from thefollowing
weak variant of Church’s thesisCTo :
V. Lifshits has sown
(in 1974, unpublished)
that even thepositive
versionCT !
of Iis
strictly
weaker(in G )
thanCT 0 .
2. CHARACTERIZATION OF (WEAKLY) RECURSIVELY INSEPARABLE PAIRS OF R.E. SETS VIA RED.
2.1. A FINITE AXIOMATIZATION OF KLEENE’S PREDICATE T.
In the
language of ~ l, fix
threepredicates Eq(x,y), Z(x)
andS(x,y).
We think ofthese as
representing equality,
zero and the successor relation. LetAS ( *
the axiom of thetheory
ofsuccessor)
be theconjunction
of the closure of thefollowing
formulaeConsider the familiar
definig
schemata forprimitive
recursive functions :These
equations
may be axiomatizedby
The characteristic function T of Kleene’s
predicate
T isprimitive recursive ;
there is therefore a liste
~~~,e.
of
equations,
each of which is an instance of one of(1) - (5)
with someauxiliary
functionletters
f ,
..., t , which defines T .Taking
thecorresponding
instances of(1) * - (5) *,
with
predicate
lettersF ,
0 ...,Fk,
Tcorresponding
tof ,
0 ..., t , we obtain a listB ,
...,Bm
of formulae which defines T. We letwhere
Ci
is the closure ofBi (i
Lm).
Let z ,..., zi, ...
be a list of variables(not occurring
0 in
A) ;
define2.2. We refer below to the schema
Write
(in
thelanguage
of £2)
where W e
is the number theoreticpredicate
o
Val w, E1 (rG(a1,a2)7)
then reads : the r.e. setsW a 1
andWa2
are notseparable by
anypair
of decidable r.e. sets.
2.3. LEMMA.
(in ~~)
where n : =
max [ a,b,c ].
PROOF : One proves
for each
primitive
recursivepredicate Fi
used in the definition of T,by
induction on thelength
of definition ofFi.
The details are routine. 02.4. LEMMA.
(in £ 2)
Hence
where Z, ..., T are the number theoretic
predicates
for zero, successor etc.Since the
premise
of(1)
istrivially
true, we obtainVal~
asrequired.
II. Assume
Fix P and
Q
and assumeand
From
(2)
and(4)
we thenget
Assume
which, by
lemma 2.3. andassumption (3) implies E [ T,P,Q ] (zm,zn).
Thisbeing
derived from(6),
weget by
intuitionisticpropositional logic
that(5) implies
-~2013 E[ T,P,Q ] (z , z~)
as
required. o
2.5. LEMMA.
(in £ 2>
REDimplies
PROOF : The
implication
from left toright
is trivial. Assume on the other handand, fixing
P andQ,
assume D[P,Q ]. By
RED D[ P,Q] implies
the weak existence of some el,e2
s.t.which
by (1) implies ,IE [1 ,P,Q] I (m,n)
asrequired. 0
2,6. PROPOSITION,
(in - 2)
REDimplies
PROOF : Immediate from 2.4 and 2.5. ~
3. WEAK INCOMPLETENESS
OF I ,
(UNDER RED).,
/
w,E01
3. 1. PROPOSITION. The
species
S := m,n) ( Val 1
is not r.e.PROOF : Assume that the
speciesS
is r.e., i.e. - for someprimitive
recursivepredicate
lI.Let
where
Note that we may define a
primitive
recursiveproof predicate Prov C+
forG as
follows :where
imp
is aprimitive
recursive function which satisfiesand where
Prov ~
iseasily proved
in G tosatisfy
theelementary derivability
conditions. Let Gwhere neg is a
primitive
recursive function which satisfiesneg( r F ’ ) = ~ 2013!F .
for any formula F in thelanguage
SinceG
is r.e., theconsistency of G is equivalent
tothe statement
oWa 1 and Wa 2 are recursively inseparable» (see Smullyan [ 61 ]
p. 6~. thm.27 ;
notice that theproof
there isintuitionistic).
Put
formally :
where Con + : = -
Pr +(r_l_7).
In£2
we know however that Con + is true,20132013C G 2 G
and
consequently Va l’ a 2
must be an axiomof Ct,+. Together
with(2)
thisimplies
Con - +
which contradicts G6del second
incompleteness
theorem(since Prov ,
satisfies theelementary
derivability conditions).
Thus theassumption
that S is r.e. leads2)
to a contradiction. 8 REMARK : The reference toSmullyan’s
theorem is made for the sake ofbrevity.
One may instead
pick
up anypair We , We
1 2of r.e. sets which areproved
in G to berecursively inseparable (cf.
e.g.Rogers [ 67 1
p. 94 thm.XII(c),
whoseproof
isintuitionistic).
Defining
we find that
Con
isequivalent (in C)
to the recursiveinseparability
ofWd
andWd2.
C
1 2THEOREM I.
(Weak incompleteness).
InL2
+ RED we can prove that not every valid formula isprovable.
PROOF : If every valid formula of
£, 1
isprovable
in ~ i then thespecies
of valid formulae is the same as thespecies
ofprovable formulae,
and so it must be r.e.. The formulae of the formAn - G(zm, zn)
arerecursively recognizable,
and thus the valid formulae of this form make upan r.e.
species
as well.By
2.6. S is then r.e.,contradicting proposition
:t 1. above.3.2. The result of 3.1 can be
classically improved by
thefollowing
PROPOSITION :
Wm, Wn rec. inseparable)
isII2-complete.
oPROOF.
(essentially
due to C.Jockusch)
Fix apair Rl, R’2
ofrecursively inseparable
r.e. sets.Let
k(e), hl(e), h2(e)
beprim.
rec. functions defined(through
the s-m-ntheorem) by
Then :
(i) We
finite(ii) We
infiniteSo the set
reduces to S. But I is known to be
II02-complete (cf. Rogers [67] p. 326, 1.3),
and hence0
S is also
n 2-complete. 8
3.3. PROPOSITION. There are numbers m,n for which we can prove + RED that
An "4 G(zm, zn) is
valid but notprovable in ¿ 1.
P ROOF :
By
theproof
of 3.1. there are m, n s.t. and soby
2.6An ~ G(zm, zn)
is
valid, assuming
RED. On the otherhand,
ifthen
(1) ~ G[
T,P,Q] (m,n)
for any unary number theoreticpredicates
P,Q.The G6del translation
Go
of G(see
Kleene[ 52 ]
p.493)
is therefore alsoprovable
inC ;
but Do is
provable (idem, p.119 *51 a),
andby
intuitionisticpredicate logic E[ T,P,Q]
0implies
--7 --iE[
T,P,Qwhenever
P and Q do notcontain Y ,
so(1) implies
which
implies (in ~ 2)
that -T E[ T,WM,Wn] (m,n)
is true.For
the m,n consideredWm
and
wn
are however(provably) disjoint,
and soby
the definition of E we have -,E[T,Wm lwn (m,n),
a contradiction. DREMARK : A
simpler proof
of the aboveproposition
runs as follows.Classically
the schema G isequivalent
to E. E isobviously
not validclassically,
so it cannot beprovable
even in classicalfirst order
logic, by
G6del’scompleteness
theorem. The formalization of thisproof
in theintuitionistic
theory
ofspecies
is howeverslightly problematic.
3.4. INCOMPLETENESS W.R.T. VALIDITY WITH CHOICE PARAMETERS
Let a be a variable for choice sequences, and assume that for the kind of choice sequences considered we have
which is the case for the choice sequences
investigated by
Kreisel-Troelstrar
70] (cf.
6.2.1there).
(1) implies quite trivially
for everynegated
formula-,A(a )
Let us refer now to a notion of
validity
with choiceparameters,
where
Ga
comes from Gby replacing
each atomic subformulaP(tl,
...,tn) by
P(a , tl,
...,tn)- (This
is not weaker thanallowing
the choiceparameters
to be distinct for eachpredicate letter,
as canreadily
beseen.)
A
straightforward
observation shows now that the discussion aboveremains
correctcs w,CS w
when Val and
Valw
arereplaced by
Val and ananalogue Val w,GS respectively,
Val 1remains
unchanged
and the schema RED isgeneralized
toBut
by (2) REDCS
isimplied outright by
RED, since RED isnegative,
and(~-3)
with avarying
over total recursive functions isjust
aspecial
case of(the quantified
variantof)
RED.To
recapitulate,
we have obtainedTHEOREM II. In
~2
+ RED we can prove that thespecies
of formulae(of ~:1)
which arevalid with choice
parameters
is not r.e., and we may exhibit aspecific
formula of 14 1 which is valid with choiceparameters
but notprovable in
4. ~ 1
IS NOT4.1. DEFINITION OF MAXIMALITY.
Let C be a class of number theoretic
predicates
and. be a formalsystem
in alanguage extending
thelanguage
of arithmetic. A formulaH[ Po,
...,Pk J
of11
is aC-schemaof 3 when
for every
Po-
I ..., ...,* pk in C. We define formally for 8 = a,C = the class of E
in C . We defineformally
for ’)= G ,
C = the classof 2
Ipredicates :
where (for
m~1)
(In defining
thepredicate
Sch weimplicitly
use aprimitive
recursivesubstitution function, the
definition of which isroutine).
A
system L
oflogic
is said to be C -maximalfor a
if~ = ~ H ( H
in thelanguage His -
C -schema ofS . }
de
Jongh
hasproved (in 1969)
that the intuitionisticpropositional logic
isE01-maximal
for intuitionisticarithmetic G (and
certain extensions ofC). 141
Leivant1
1
(chs. B3-B5)
it isproved
that£1
is IT2-maximal
for G(and
certain extensionsof C ).
We shall now show that - is 1
not E01-maximal
1, even for afragment of G .
.4.2, For a schema H =. H
[Z,S,Eq,F 0’
...,Frg T ; P
...,Pt ] Of 1 we define
theschema
G ,
and the schemaof L1G,
where Z, ..., T are the intended number theoreticpredicates,
i.e. -Z(x) :
= x =0,
etc.If the r.e.
interpretations
of the zero, successor and
equality predicates satisfy
the axiom of successorAS (cf. 2.1),
thenZ ~
andS ~ generate
a structureisomorphic
tothrough
a recursive(total) function,
defined as follows :We let then
For a formula H of G we now write
for
the sdwmaof jiG
I 1I1I1t’. II I,
by restricting
allquantifiers
toN d’
4.3. I,EMMA.
(in G )
Letwhere n : = defined as in 2.], but with defining formulae for all 1
predicates TJ , where j
ranges over the dimensions)) of thepre()1(’" If’
iii I I .PROOF : Let G be given by a Gentzen natural deduction system (rl.e.g
PraBBit7 ! 7! !).ForaformulaH =
I (n
0...,nm) of
G forH [Z,S,Eq,..., T] (zno, ...znm]
as defined in 4.2. above. I.e. -It is easy to
verify
that for each inference rule of the (;entzen ~) ...1, 111(where
n , ...,np
is the list of all numeralsoccurring
in the formulae shown) the0 P
isaderivedruleofG+
An d, where
n:= max Pif a formula 11(ii0 , ,...,n )of
Gis derivable by a natural deduction A
then the tormnh0 P
H[d I Id (zn 0
occurring
in A .By
the existentialconjuncts
of the axiom Xs t, (.). and (B) in 2 i q may be cut down to n : :=~ maxI
i ~pl.
.Assume now the
premise
of(1),
and fix d and e.
For j -
t let thepredicate
letterp.
be In.. IpLH’1
i iiil d,’l’íllt iliiilil.t IJ J
e*= ((e*) 0 ,
0 ...,(e*)t > through
the s-m-n theoremby
where the
function v ,
whichdepends
ond,
is defined as in 4.2. above.Further,
let e 1= « e 1) o
0,..., (e 1 )t)
be definedby
By (2)
nowand so
by
the aboveargument
where n : = max
[nj|j
J pJ .
It isreadily
seen however that forwhere
for j ~
tHere
T d
is thed-interpretation
of thepredicate Ti
i.e. -T d :
zW(d)a
for some q.by
the definition of theinterpretation generated by d, provided A n d
holdsSince in H
d ~ all quantifiers
are indeed restricted toN ,
we thus have from(6)
(provided And ),
and so(5) implies
theprovability
in Ci ofThis
being
the case forarbitrary
d and e we have obtained the concliisinn of(1 ),
as4.4. THEOREM III.
£. 1
isnot E01-maximal for G . Specifically -
are m and n f’or whiuhAT, 1.4 E [ T,P,Q ] zn)
isa 2: 0 -schema
ofC ,
but is notprovable iii Y, 1. (Here
theschema E is defined as in
2.2.).
PROOF : Let
W~, Wn
be apair
of r.e. sets,proved
in G to berecursively inseparable (as
inI.e. - for each
e0, el
which
by
lemma 4.3implies
that for anyd,e
But E is
existential,
and thereforeI implies (in predicate
tB. So(1)
forarbitrary
o
d,e actually
states thatAn -> E(zm, zn) 01-schema
of L .On the other hand this formula cannot he
proved ~n
as vBc seen inREFERENCES
KLEENE,
S.C.[52],
Introduction toMetamathematics, Wolters-Noordhoff, Groningen, 1952.
[69], Formalized
Recursive Functionals and FormalizedRealizability,
Memoirs of the AMS 89
(1969).
KREISEL,
G.[70],
Church’s thesis : a kind ofreducibility
axiom of constructivemathematics ;
in lntuitionism and
Proof Theory. (eds. Kino, Myhill, Vesley) (North Holland, Amsterdam, 1970),
pp. 121-150.KREISEL,
G. and TROELSTRA, A.S.[70],
Formalsystems
for some branches of intuitionisticanalysis ;
Annalsof
MathematicalLogic
1(1970),
pp. 229-387.LEIVANT,
D.[75], Absoluteness of
IntuitionisticLogic ;
PhDdissertation, University
ofAmsterdam
(Mathematisch Centrum, Amsterdam, 1975).
PRAWITZ,
D.[71 ],
Ideas and results of ProofTheory ;
inProceedings of
the Second ScandinavianLogic Symposium (ed. Fenstad) (North Holland,
Amsterdam1971),
pp. 235-307.ROGERS,
H.[67],
TheTheory of
Recursive Functions andEffective Computability (McGraw-Hill,
NewYork, 1967).
TARSKI,
A.[36] ,
DerWahrheitsbegriff
in denformalisierten Sprachen,
Studia Phil.1(1936)
261-405.